Not really true. Texas Hold 'Em bots deal with plenty of uncertainty and can supposedly beat the top players at their game. I really just think it’s a matter of the amount of effort being put into developing a Stratego AI.
An experienced player reaches towards a bomb, hesitates as if considering moving it, then does something else.
(this usually only works once).
The computer is not among champion level at Stratego, and its calculations work best with known facts.
Why do you think there is no world class AI? Isn’t that an excuse in the year 2017?
Its because a computer, the way they work is with 1’s and 0’s, but in this game you get none of the above at the start.
Only unknowns. Real time strategy is something a computer can not do. If they could, we would not need to program them at all.
Maybe a new player would fall for that.
I once won a game down several pieces by reaching toward my flag, which did make the opponent think it was not the flag!
Later I had a successful all or nothing attack with 2 miners and my highest piece being a #3 ( He had his #1 left, with my spy gone and more overall left ) by correctly guessing where his flag was.
Work best is a far cry from cannot deal with unknowns.
Well, I already said I don’t know for certain but it appears to be because there’s not enough people interested in it. There’s very little research being on Stratego AI compared to other games. AIs don’t just port over from one problem to another so much. I do know for certain that it is not for the reasons you’re saying.
And the rest of this is just all wrong.
If you’re interested in learning about this, there’s quite a few articles on this online but they’ll likely be pretty technical if you don’t have a computer science background and more specifically an AI background. Although these topics are usually taught in an Introduction to AI course so a computer science background should be sufficient.
You can look up things like:
Uncertainty in artificial intelligence
Belief networks
Belief systems
Knowledge representation
These subjects will give you a good primer on how AI deals with uncertainty. If you want, ask me and I’ll try to find some specific readable material for you if you provide me some indication of your degree of knowledge of computer science. But that will take quite a bit of my time, so I don’t want to do that if you’re not really that interested in learning about it.
What does Stratego have to do with real-time strategy? Whatever else it may be, Stratego is turn-based, same as chess or go.
And there has been some work on AIs for real-time strategy, too, and they beat humans at that, too.
You don’t know for certain, like you said. And how do you do how many people are playing it.
Beep, can you keep an open mind on this topic?
Read the below link, page three please.
**" What Satz has done with his Stratego-playing AI, Probe, is program it to know human tendencies. It can’t read your poker face, but it knows the situations in which humans are statistically most likely to bluff; for example, it knows that players often act aggressively with a weak piece to try to mislead the opponent. It knows—from watching humans play, and because Satz told it so—that most players prefer to have strength on their right sides rather than their left (perhaps an effect of our species’ dominant right-handedness, a quirk computers do not share). And, most dangerously, it remembers what you like to do and how you like to arrange your pieces. “Some people have played several thousand games against it,” Satz says. “Probe is able to remember all the setups those opponents used, which is devastatingly strong information.”
Even so, Probe—which dominates other Stratego AIs in computer versus computer tournaments—can match the skill of only average human player! The AI challenge of dealing with incomplete information is too great.**
http://discovermagazine.com/2011/jan-feb/14-whos-smarter-human-computer-round-9-jeopardy
>>> You might also want to know AI computers can beat the better human opponents int he game of Risk. I play risk, and the AI sucks there too, but admittedly is was only on playstation’s AI.
In 2011? That’s an eon ago in AI. In 2011 the best computer Go players couldn’t beat high ranked amateurs. Now it’s beating top Pros.
Ok, I read it. Fairly interesting for a pop-sci piece.
Have you read anything on the subjects I suggested above? Here’s just a small sample of work done on uncertainty, much of it since 2013.
[1] Lemmer, J. F., & Kanal, L. N. (2014, June). Propagating uncertainty in Bayesian networks by probabilistic logic sampling. In Uncertainty in artificial intelligence (Vol. 2, p. 149).
[2] Cohen, P. R., & Feigenbaum, E. A. (Eds.). (2014). The handbook of artificial intelligence (Vol. 3). Butterworth-Heinemann.
[3] Wenger, E. (2014). Artificial intelligence and tutoring systems: computational and cognitive approaches to the communication of knowledge. Morgan Kaufmann.
[4] Ghahramani, Z. (2015). Probabilistic machine learning and artificial intelligence. Nature, 521(7553), 452-459.
[5] Horvitz, Eric J., and David Heckerman. “Modular belief updates and confusion about measures of certainty in artificial intelligence research.” arXiv preprint arXiv:1407.7281 (2014).
[6] Gabbay, D. M., & Smets, P. (Eds.). (2013). Quantified Representation of Uncertainty and Imprecision (Vol. 1). Springer Science & Business Media.
[7] Castillo, L., López, C., Bedia, M., & López, F. (2015). Collective Strategies for Virtual Environments: Modelling Ways to Deal with the Uncertainty. In Distributed Computing and Artificial Intelligence, 12th International Conference (pp. 183-191). Springer, Cham.
[8] Dubois, D. J., Wellman, M. P., & D’Ambrosio, B. (Eds.). (2014). Uncertainty in Artificial Intelligence: Proceedings of the Eighth Conference (1992), July 17–19, 1992, Eighth Conference on Uncertainty in Artificial Intelligence, Stanford University. Morgan Kaufmann.
[9] Millington, I., & Funge, J. (2016). Artificial intelligence for games. CRC Press.
[10] Webber, B. L., & Nilsson, N. J. (Eds.). (2014). Readings in artificial intelligence. Morgan Kaufmann.
[11] Beierle, C. (2017). Management of uncertainty in Artificial Intelligence and databases.
[12] Spaan, M. T., Veiga, T. S., & Lima, P. U. (2015). Decision-theoretic planning under uncertainty with information rewards for active cooperative perception. Autonomous Agents and Multi-Agent Systems, 29(6), 1157-1185.
[13] Bench-Capon, T. J. (2014). Knowledge representation: an approach to artificial intelligence (Vol. 32). Elsevier.
[14] Sigaud, O., & Buffet, O. (Eds.). (2013). Markov decision processes in artificial intelligence. John Wiley & Sons.
[16] Shachter, R. D., & Peot, M. A. (2013). Simulation approaches to general probabilistic inference on belief networks. arXiv preprint arXiv:1304.1526.
[17] Cooper, G. F. (2013). A method for using belief networks as influence diagrams. arXiv preprint arXiv:1304.2346.
[18] Mnih, A., & Gregor, K. (2014). Neural variational inference and learning in belief networks. arXiv preprint arXiv:1402.0030.
[19] Shachter, R. D., Kanal, L. N., Henrion, M., & Lemmer, J. F. (Eds.). (2017). Uncertainty in Artificial Intelligence 5 (Vol. 10). Elsevier.
[20] Goldman, R. P., & Charniak, E. (2013). Dynamic construction of belief networks. arXiv preprint arXiv:1304.1092.
[21] Wen, W. X. (2013). Optimal decomposition of belief networks. arXiv preprint arXiv:1304.1111.
[22] Gabbay, D. M., & Smets, P. (Eds.). (2013). Quantified Representation of Uncertainty and Imprecision (Vol. 1). Springer Science & Business Media.
And so on.
It might help you to know that I teach artificial intelligence courses at an undergraduate and graduate level. I’m also an active researcher in applications of artificial intelligence. That doesn’t mean of course that I’m aware of everything that is going on in research on AI of course, but what you’re talking about is very basic and taught in an Introduction to Artificial Intelligence course. So you’ll need to do a little bit better than a layman’s supposition and a pop-sci article.
I looked up Satz. It might interest you to know that he stopped working on Stratego. In 2011. I guess he didn’t find it that compelling either. But hey it was good enough to get him a partial interview in a pop-sci magazine, so I applaud him for that. (I’m being serious here, getting those interviews isn’t easy and getting your name out there is really good, so good for him.)
Excellent! Then you are in a perfect position to find out why AI can’t match people on games like Stratego!
I’m sure the algorithms and programming are impressive for these top computer AI gamers, but what happens to them when you take away their data? Well, that is how the game of Stratego starts.
Below is taken from a white paper of not one but several Computer AI’s in their own Stratego tournament.
Until Machines can think for themselves, they won’t dominate non recall or math type of games where strategy, intuition, and the unknown matter. An EQ, not an IQ if you will.
" While the overall strength of programs has progressed, it is unlikely that they can match the human skill within the next few years (Schadd and Satz, 2009). However, it is the hope of the programmers that a common body of knowledge will evolve out of the growing number of tournament participants and that one-day human’s players will see a serious challenge in competing with a Stratego AI. Computers, in general, have grown above and beyond in the last two decades, therefore nothing is impossible. The computer championship has been an upbeat event and hopes to continue to attract interest from programmers all around the globe. "
http://www.schadd.com/Papers/2009StrategoWorldChampionship.pdf
Before determining why something is true, you should first determine whether it is true. You have not yet done so. No current AI can beat the top humans at Stratego, but that does not imply that AI can’t do it.
And again, computers are already better than the best humans at poker, another game of incomplete knowledge, bluffing, strategy, and intuition. If, as you claim, computers are unable to play such games, then how do you explain that fact?
Lack of research interest would be the most likely explanation.
Well, you can keep repeating the same incorrect thing over and over but it won’t make it more correct. There are numerous algorithms for dealing with uncertainty.
Like poker? So, again, you can keep repeating the same incorrect thing over and over, but that won’t make it more correct. There are numerous algorithms that deal with unknowns. Here’s an introduction that you won’t read.
Since, you seem resistant to learning and it is my policy not to simply go around in circles, I’m done trying to convince you of the obvious. I simply don’t have the time to waste on such endeavors. I love teaching but not to people who don’t care to learn. If you are ever interested in actually learning about how AIs deal with uncertainty, you can start with the introduction above and if you PM me I’ll be more than happy to direct you to additional material on the matter.
So what? And what does this even mean, really?
Suppose I find someone, Alice, who’s never heard of chess. I teach her a game, schpless, which happens to have exactly the same rules as chess, and she gets very good at it.
Alice doesn’t know she’s playing chess. And yet she plays chess perfectly well. Does this mean anything? Could it mean anything? No.
Even for ordinary human chess-players, the part of the brain they’re exercising largely doesn’t know it’s playing chess. They’re overloading a general-purpose pattern recognizer that evolved to detect tigers or antelope droppings or whatever. A huge fraction of it is totally subconscious. And the part that is conscious turns out to be irrelevant. Which may mean there’s no meaningful distinction there, either.
Says who? Would you want your AI to win and be champion of this game too? They can’t, and there are several…
Now your mixing apples with oranges. Poker is not Stratego. If you have to do that to make your point, what does it really say? AI can not think for itself. It never could. In a game with many unknowns, all the match and algorithms in the world haven’t helped much.
Learn, sure. I can learn, but you seem to mentally block the data I give you that says AI is a loser vs. a good mid level human player. I have data, you have none, except to lean upon unknowns.
Says literature. There’s been no work on it since 2011. This indicates a lack of interest. If you want me to answer for what’s going on in the minds of AI researchers around the world, I’m afraid I cannot do that. It could be a lack of funding. It could be difficulty in getting papers published. The interest in Stratego was very brief, so it could be that it was the focus of some student’s thesis and now that he’s graduated the interest is gone. What I can say with absolute certainty is that it is not because AI cannot deal with uncertainty.
Your problem here is that everything is recorded. So let’s review shall we?
[QUOTE=BeepKillBeep]
This is absolutely not true. There are numerous algorithms that deal with uncertainty. It might very well be true that there’s no world class Stratego AI, I have no idea, but it isn’t because of that.
[/quote]
(bolding mine)
So let’s address what might be your sticking point. You seem fixated on that the best Stratego AIs aren’t that good against even a mid-level player. I’ve never addressed that other than to say I don’t know one way or the other. It seems to be a case of lack of research interest with no work done on Stratego since 2011 as discussed above.
What I have attempted to address is your repeated errors that AIs cannot deal with unknowns or uncertainties. This is factually incorrect. Why you continue to focus on the non-existence of a world class Stratego AI instead of correcting your obvious factual errors, I cannot explain. Well, I can but not in this forum.
Bye bye.
Of course I would. So?
For perspective: I once wrote a program that plays Connect 4. It’s… Well, it’s OK. It can beat most children, and beats me about a quarter of the time, when I make a sloppy mistake. Now, I could make it better, and even have a clear idea about how I could do it. And in fact, I know it’s possible to write a perfect Connect 4 AI, because it has in fact been done by others. So why haven’t I done it? Eh, I just haven’t bothered.
Let me reiterate that: The reason my Connect 4 game doesn’t play perfectly (or at least, a lot better than it does now) is not because it’s impossible. It’s just that I lost interest in the project after the first version. It is highly likely that the lack of world-champion-quality Stratego players is for the exact same reason.
Dare I point out that a world-champion-quality Stratego AI need not play a perfect game, nor indeed play very well. All it has to do is beat puny humans, which is not saying very much 
As has been pointed out above, there is interest in general game-playing AIs that learn and improve by playing many times, and this does work, but like the others I do not know of any recent research specifically involving Stratego.
AI has been around for decades. The first big news is when IBM’s Deep Blue defeated world chess champion, Garry Kasparov. This happened in 1997.
My point, which you may have an issue with is several AI’s many years after 1997 can not beat good mid-level human players at Stratego. Not championship level player, mind you.
When you can’t win, some attack the sources provided, but this becomes problematic when the attacker has no sources at all to offer the contrary. How do you know the AI programmers themselves felt it wasn’t going to work and pulled their own plugs? You don’t. At least I have provided sources on the topic.
You have to ask, why can’t the AI win in this game? As I failed to convince you, Chess and Stratego are very different One if perfect for AI and it’s as transparent as can be, the other with pieces of unknown quantities where all the math and probability in some cases will not help.
Since I am a decent player at both Chess and Stratego maybe it’s my fault to explain to him how Stratego is very different. A good mid-level player in Stratego, if he plays with a lot of risks can beat an expert! The best in the world. If you can understand this, you have the ability to change your opinion on AI upside in this game.
This is not the case at chess at all where a high master level would beat a good mid level player 99 times out of 100, and the best AI would beat a mid level player 999,999 times out of 1,000,000
So you see there is no such thing as an un-beatable opponent in the Stratego unless someone is cheating. Anyone, if they are somewhat familiar with the game, can launch a heavy assault to the left middle or right, and " guess " where the flag is.
Now that you understand the above, do you still think the AI will become unbeatable in Stratego?
I just want a yes or no out of you on the the bolded as the reply will tell me everything I need to know.